目录

如何理解系统的可观测性

目录

系统的可观测性的定义为:

对于系统 https://www.zhihu.com/equation?tex=%5Cbegin%7Baligned%7D+%5Cdot%7Bx%7D%28t%29+%26%3DA%28t%29+x%28t%29+%5C%5C+y%28t%29+%26%3DC%28t%29+x%28t%29+%5Cend%7Baligned%7D 在有限的时间区间里 https://www.zhihu.com/equation?tex=%5Bt_0%2Ct_%7B%5Calpha%7D%5D 内,对应初态 https://www.zhihu.com/equation?tex=x%28t_0%29+%3D+%5Cbar%7Bx%7D ,有

https://www.zhihu.com/equation?tex=y%28t%29+%5Cequiv+0%2C+%5Cquad+t+%5Cin%5Cleft%5Bt_%7B0%7D%2C+t_%7B%5Calpha%7D%5Cright%5D 则称为不可观测状态,如果不存在这样的不可观测状态,那么系统完全可观测。

这个定义说的非常的抽象,但是可观测性这个概念最重要的是其在 状态观测器(state observer)的设计中发挥了关键作用,下面我就从这个角度进行一些对于状态可观性的介绍。


其实,状态可观性的概念是出自于卡尔曼,而著名的Kalman filter也实质上是state estimator for stochastic system[1]

对于系统进行控制,首先我们需要确定要进行的控制的类型,一般而言,控制分为基于状态变量的控制和基于输出的控制,这其中最好的当然是状态控制,但是我们往往只有输出和输入,而难以直接得到系统的状态变量,这个时候就需要从前面所说的输出和输入还原出系统的状态变量,这就是所谓的状态观测[2]

https://pic4.zhimg.com/80/v2-060801e2b7592a23a3828b7046713790_1440w.jpg状态反馈和输出反馈

https://pic3.zhimg.com/80/v2-3266d61a72e80409f698d100d13f1805_1440w.jpg典型状态观测器结构


一言以蔽之,系统状态的可观测性是进行状态观测器设计的前提,也就是说,无论是用什么方法,只有当系统完全可观时,才有可能从状态观测器的设计得出系统的状态的观测。

以Full-Order Observers为例[3],其结构如下:

https://pic3.zhimg.com/80/v2-930737553d9c84c4b6ccccbb681cedc3_1440w.jpg

对于观测系统有: https://www.zhihu.com/equation?tex=%5Cdot%7B%5Chat%7Bx%7D%7D+%3D+A+%5Chat%7Bx%7D+%3D+Bu+%5Cquad+%5Chat%7By%7D+%3D++C+%5Chat%7Bx%7D

跟踪误差为: https://www.zhihu.com/equation?tex=%5Cwidetilde%7By%7D%3Dy-%5Chat%7By%7D%3DC+x-C+%5Chat%7Bx%7D

即: https://www.zhihu.com/equation?tex=%5Cdot%7B%5Chat%7Bx%7D%7D%3DA+%5Chat%7Bx%7D%2BB+u%2BL%28y-%5Chat%7By%7D%29 和原始系统做差有:

https://www.zhihu.com/equation?tex=%5Cbegin%7Baligned%7D+%5Cdot%7B%5Cboldsymbol%7Bx%7D%7D-%5Cdot%7B%5Chat%7B%5Cboldsymbol%7Bx%7D%7D%7D+%26%3D%5Cboldsymbol%7BA%7D%28%5Cboldsymbol%7Bx%7D-%5Chat%7B%5Cboldsymbol%7Bx%7D%7D%29-%5Cboldsymbol%7BL%7D%28%5Cboldsymbol%7By%7D-%5Chat%7B%5Cboldsymbol%7By%7D%7D%29+%5C%5C+%26%3D%5Cboldsymbol%7BA%7D%28%5Cboldsymbol%7Bx%7D-%5Chat%7B%5Cboldsymbol%7Bx%7D%7D%29-%5Cboldsymbol%7BL%7D+%5Cboldsymbol%7BC%7D%28%5Cboldsymbol%7Bx%7D-%5Chat%7B%5Cboldsymbol%7Bx%7D%7D%29+%5C%5C+%26%3D%28%5Cboldsymbol%7BA%7D-%5Cboldsymbol%7BL%7D+%5Cboldsymbol%7BC%7D%29%28%5Cboldsymbol%7Bx%7D-%5Chat%7B%5Cboldsymbol%7Bx%7D%7D%29+%5Cend%7Baligned%7D

跟踪误差可以求得: https://www.zhihu.com/equation?tex=%5Cboldsymbol%7Bx%7D%28t%29-%5Chat%7B%5Cboldsymbol%7Bx%7D%7D%28t%29%3D%5Cmathrm%7Be%7D%5E%7B%28%5Cboldsymbol%7BA%7D-%5Cboldsymbol%7BL%7D+%5Cboldsymbol%7BC%7D%29%5Cleft%28t-t_%7B0%7D%5Cright%29%7D%5Cleft%5B%5Cboldsymbol%7Bx%7D%5Cleft%28t_%7B0%7D%5Cright%29-%5Chat%7B%5Cboldsymbol%7Bx%7D%7D%5Cleft%28t_%7B0%7D%5Cright%29%5Cright%5D

写成矩阵形式:

https://www.zhihu.com/equation?tex=%5Cleft%5B+%5Cbegin%7Barray%7D%7Bc%7D%7B%5Cdot%7B%5Cboldsymbol%7Bx%7D%7D%7D+%5C%5C+%7B%5Cdot%7B%5Cboldsymbol%7Bx%7D%7D-%5Cdot%7B%5Chat%7B%5Cboldsymbol%7Bx%7D%7D%7D%7D%5Cend%7Barray%7D%5Cright%5D%3D%5Cleft%5B+%5Cbegin%7Barray%7D%7Bcc%7D%7B%5Cboldsymbol%7BA%7D%7D+%26+%7B0%7D+%5C%5C+%7B0%7D+%26+%7B%5Cboldsymbol%7BA%7D-%5Cboldsymbol%7BL+C%7D%7D%5Cend%7Barray%7D%5Cright%5D+%5Cleft%5B+%5Cbegin%7Barray%7D%7Bc%7D%7B%5Cboldsymbol%7Bx%7D%7D+%5C%5C+%7B%5Cboldsymbol%7Bx%7D-%5Chat%7B%5Cboldsymbol%7Bx%7D%7D%7D%5Cend%7Barray%7D%5Cright%5D%2B%5Cleft%5B+%5Cbegin%7Barray%7D%7Bl%7D%7B%5Cboldsymbol%7BB%7D%7D+%5C%5C+%7B0%7D%5Cend%7Barray%7D%5Cright%5D+%5Cboldsymbol%7Bu%7D

我们希望跟踪误差趋近于0并且其衰减速度可以调控,显然这取决于矩阵 https://www.zhihu.com/equation?tex=A-LC 的特征根,这就是所谓的极点配置的问题:

  • 选择一个合适的 https://www.zhihu.com/equation?tex=L ,使得 https://www.zhihu.com/equation?tex=%5Coperatorname%7BRe%7D%5C%7B%5Clambda%28%5Cboldsymbol%7BA%7D-%5Cboldsymbol%7BL%7D+%5Cboldsymbol%7BC%7D%29%5C%7D%3C0+%5Cquad+%5CLongrightarrow+%5Cquad+%5Clim+_%7Bt+%5Crightarrow+%5Cinfty%7D%5Bx%28t%29-%5Chat%7B%5Cboldsymbol%7Bx%7D%7D%28t%29%5D%3D0
  • 如果 https://www.zhihu.com/equation?tex=%5Coperatorname%7BRe%7D%5C%7B%5Clambda%28A-L+C%29%5C%7D%3C-%5Csigma 就可以保证所有的初始值都可以按快于 https://www.zhihu.com/equation?tex=%5Cmathrm%7Be%7D%5E%7B-%5Csigma+t%7D 的速度收敛

而根据对偶原理,对于矩阵 https://www.zhihu.com/equation?tex=A-LC 的极点配置的问题相当于利用状态反馈 https://www.zhihu.com/equation?tex=L%5ET 对于系统 https://www.zhihu.com/equation?tex=%28A%5ET%2CC%5ET%29 的极点配置问题,而这等价于系统 https://www.zhihu.com/equation?tex=%28A%5ET%2CC%5ET%29 可控,那么即等价于系统 https://www.zhihu.com/equation?tex=%28A%2CC%29 可观。这里就解释了可观和可控的对偶的用处以及系统可观性的实际含义。

参考

  1. ^段广仁. 线性系统理论. 哈尔滨工业大学出版社, 1996 https://item.jd.com/30834945675.html
  2. ^吴麒主编. 自动控制原理 (第2版, 下册). 清华大学出版社, 2006. http://www.tup.tsinghua.edu.cn/booksCenter/book_02284603.html
  3. ^Discrete Control System, Katsuhiko Ogata https://www.goodreads.com/book/show/241242.Discrete_Time_Control_Systems